Agent Utility Evaluation on ASB 5 direct prompt-injection styles
88.9Utility (%)SecureClaw
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SecureClawBase Model=gpt-4o-mini-2024-07-18, Temperature=0, Interface=bounded-summary interface2026.06 | 88.9 | 1,778 | 87.44 | — | |
| FarameshBase Model=gpt-4o-mini-2024-07-18, Temperature=0, Modification scope=Core runtime unchanged; harness supplies adapters2026.06 | 3.7 | 74 | 2.92 | 10 |